A Soft Handover for Service Delivery in Intermittently ... · A Soft Handover for Service Delivery in Intermittently Connected Hybrid Networks 3 according to the quality of the multi-hop

A Soft Handover for Service Delivery in IntermittentlyConnected Hybrid Networks

Nicolas Le Sommer, Ali Makke and Yves Mahéo

IRISA Laboratory, Université de Bretagne Sud, France{Nicolas.Le-Sommer, Ali.Makke, Yves.Maheo}@univ-ubs.fr

Summary. Today, handheld devices equipped with Wi-Fi interfaces are used in-tensively by a huge number of people every day. These devices can form inter-mittently connected mobile ad hoc networks spontaneously. These networks ap-pear as a relevant solution to extend a pre-existing infrastructure-based networkcomposed of several access points in view of providing nomadic people withapplication services in a wide area. In such hybrid networks, intermittent connec-tions are prevalent, and end-to-end paths between clients and providers cannot bemaintained all the time. Thus, the communications must be achieved following a”store, carry and forward” principle.In this paper, we present a new soft handover mechanism dedicated to servicedelivery in such hybrid networks. This handover solution exploits several piecesof information, such as the message propagation time, the path stability, and themobility degree of intermediate nodes in order to select the most appropriateaccess point(s) to forward a response to a given mobile client.

1 Introduction

The recent market researches on mobile computing devices show an incredible pene-tration of handheld devices equipped with IEEE 802.11 interfaces (e.g., smart-phones,internet tablets) among the population, as well as a significant growth of computing de-vices embedded in the environment (e.g., Wi-Fi access points, wireless DSL gateways,sensors). These embedded devices are irregularly, and sometimes sparsely, distributedin the environment, and are connected to different infrastructure based networks. In or-der to access to the Internet or to get some services, the mobile clients must be in thecommunication range of an access point (see Figure 1a), thus constraining their mobil-ity and reducing the area where a service or an Internet access can be offered. Over thelast years, wireless mobile ad hoc networks have been considered in order to providemulti-hop communication between devices, and sometimes to create hybrid networksby extending fixed infrastructures in order to better satisfy the user’s needs while usingfewer access points to cover a given area. The access points participate in ad hoc com-munication and provide access to the fixed infrastructure (see Figure 1b). Nevertheless,such hybrid networks still remain rarely used today because their topology suffers fromunpredictable changes and connectivity disruptions due to the mobility of the devicesand the short communication range of their wireless interfaces. These changes are alsothe result of the volatility of the mobile terminals that are frequently switched off dueto their limited power budget. In these conditions, it is difficult, and even impossible,

2 Nicolas Le Sommer, Ali Makke and Yves Mahéo

to maintain an end-to-end path between two devices using legacy MANET (Mobile Adhoc NETwork) routing protocols.

(a) Simple infrastructure-based wireless net-work.

(b) Hybrid wireless network.

(c) Multi-hop intermittently connected hybrid wireless networks

Fig. 1: From simple wireless networks to multi-hop intermittently connected hybridnetworks.

One of the most interesting evolutions of these hybrid networks is what we call inter-mittently connected hybrid networks (ICHN) or opportunistic hybrid networks (OHN),whose goal is to enable communications in presence of frequent and unpredictable con-nectivity disruptions. In such networks, communications rely on the ”store, carry andforward” principle, whose basic idea is to take advantage of device contact opportuni-ties to exchange messages, as well as of the device mobility so as to deliver messagesbetween the different partitions of the network. In ICHN, two devices can communi-cate even if it does not exist an end-to-end path between them. Such hybrid networkscould appear as an opportunity for service providers, such as local authorities, to pro-vide nomadic people with new ubiquitous services, without resorting to any expensiveinfrastructure, such as those provided by mobile phone operators. The fixed part ofthese hybrid networks can obviously present various topologies. For instance, the ser-vices can be provided by dedicated servers that can be accessed by the mobile devicesthrough the infostations, which act as gateways (see Figure 1c).

In this paper, we focus on the service delivery process in both the mobile and theinfrastructure parts of an ICHN, and we present the soft horizontal handover mecha-nism we have designed and implemented in the infostations in order to improve theservice delivery for nomadic people. Unlike the handover mechanism designed for cel-lular networks, the handover mechanism we propose takes the opportunistic nature ofthe communications into account. Indeed, the handover decisions are not taken accord-ing to the quality of the radio signal between a base station and a mobile client, but

A Soft Handover for Service Delivery in Intermittently Connected Hybrid Networks 3

according to the quality of the multi-hop discontinuous paths between a client and aninfostation. These paths, which can evolve dynamically according to the mobility andthe volatility of the devices, are characterized by several properties, such as their stabil-ity or their length.

The remainder of this paper is organized as follows. Section 2 presents some relatedwork focusing on handover mechanism as well as on communication and service deliv-ery in opportunistic networks. Section 3 introduces service provision issues in ICHN.Section 4 presents the handover solution we propose to improve service delivery inICHN. Section 5 shows experimental results we obtained for our handover solution.Section 6 concludes this paper with a discussion on open research directions.

2 Related work

Communications in disconnected or intermittently connected mobile ad hoc networkshave been investigated in research works dealing with delay tolerant networking, dis-ruption tolerant networking or opportunistic networking. The solutions presented inthese works are generally based on the ”store, carry and forward” principle. Some ofthem also make assumptions about the device mobility by considering that these equip-ments are carried by humans that follow social mobility patterns, and that such recurrentpatterns can be used to predict the future contacts between the devices and to deliver themessages efficiently with a limited number of copies of these messages. These methodstraditionally use a probabilistic metric, often called delivery predictability, that reflectshow a neighbor node will be able to deliver a message to its final recipient. Beforeforwarding (or sending) a message, a mobile host asks its neighbors to compute theirown delivery probability for the considered message, and then compares these probabil-ities and selects the best next hop(s) among them. This estimation can require a 1-hop,and sometimes a 2-hop, network knowledge. In the Context-Aware Routing protocol(CAR) [12], the delivery probabilities are computed using both utility functions andKalman filter prediction techniques. Propicman [13] also exploits context propertiesand the probability of nodes to meet the destination, and infers from that the deliveryprobability, but in a different way. When a node wants to send a message to anotherone, it sends to its neighbor nodes the pieces of information it knows about the destina-tion. Based on these pieces of information, the neighbor nodes compute their deliveryprobability and return it. The node that wants to send the message will send this mes-sage only on the two-hop route(s) with the highest delivery probability (this probabilitymust further be higher than its own one). Like CAR and Propicman, HiBOp [1] usescontext properties in order to compute delivery probabilities, but it uses history infor-mation in order to improve the delivery probability instead of making predictions usingKalman filters. In Prophet (Probabilistic Routing Protocol using History of Encountersand Transitivity) [9], when a node wants to send a message to another one, it will lookfor the neighbor node that has the highest amount of time encountering the destination,meaning that has the highest delivery predictability. Furthermore, this property is tran-sitive in Prophet. These protocols are designed for unicast communications. Thus, theycould probably be used for service invocation, which traditionally relies on such a com-munication paradigm, but not for service discovery, which requires in ICHN an efficient


broadcast of service discovery requests and service advertisements. Indeed, in order toavoid the broadcast storm problem and a network congestion, these messages must notbe broadcast in a blindly epidemic manner, but instead using dedicated protocols suchas OLFServ [8].

Software service provision with delay-tolerant, disruption-tolerant or opportunisticcommunications has been addressed so far in few research works [10, 7, 14, 2]. Propos-als in [10] and [2] focus on service provisioning in opportunistic networks composedsolely of mobile nodes. In [10], the authors propose content-based service discoveryand invocation solutions in order to exploit the redundancy of the services offered bythe mobile devices that can move freely (i.e., no assumptions are made regarding themobility of the devices). The protocol presented in [2] targets networks relying on socialinteractions between mobile nodes that act as both clients and providers of services. Dueto the volatility and the limited resources of the mobile devices, the number of relevantservices that can be offered by these devices is limited in comparison to those that couldbe offered in hybrid networks. Unlike in [10] and [2], the services considered in [7] areprovided by fixed infostations in limited geographical areas. In [7], mobile devices andinfostations are aware of their own location. Mobile devices can invoke remote infosta-tions thanks to an opportunistic and location-aware forwarding protocol [8]. In contrastwith the environments we consider in this paper, in [7], the infostations were not con-nected together.

The cooperation between wireless infrastructures and opportunistic networks hasbeen investigated recently in order to enhance the content delivery to mobile clients andto relieve the infrastructure [3, 4, 5, 16, 11]. In [4], Hui et al. show that opportunisticcommunications can improve the content delivery ratios significantly even in infrastruc-tures with a high access point density. Hui et al. also investigate different strategies tofind the subset of mobile devices that will lead to the greatest infection ratio by the endof a message’s lifetime [3]. In [5], Ioannidis et al. also focus on the delivery of dynamiccontent from the infrastructure to mobile subscribers that are expected to replicate itepidemically. They showed that, by supporting such epidemic exchanges and by utiliz-ing the bandwidth of connections between mobile devices, the content providers cansupport more subscribers with a lower cost. [16] targets the same objective than [3, 4],but it does not focus on social networks and does not assume preexisting knowledge ofpairwise contact probabilities. It proposes Push-and-Track, a framework that exploitsboth wide-area wireless networks (e.g., 3G or WiMax) and local-area wireless networks(e.g., Bluetooth or Wi-Fi) in order to achieve guaranteed delivery in an opportunisticnetwork while relieving the infrastructure. In [16], a subset of users will receive thecontent from the infrastructure and start propagating it epidemically; upon receivingthe content, mobile nodes send acknowledgments back to the source, thus allowing itto keep track of the delivered content and assess the opportunity of sending new copies.Service invocation issues in ICHN have been addressed recently with a reactive rout-ing protocol called TAO (Time-Aware Opportunistic Routing Protocol) [11]. In TAO,the routing decisions are taken based on the last date of contact of mobile devices withinfostations. Furthermore, TAO implements several optimizations, such as source rout-ing techniques, so as to perform an enhanced service delivery. However, in its current


version TAO does not include pieces of information that could help at taking handoverdecisions.

Several kinds of handover mechanisms and algorithms have been proposed in thepast for various types of wireless networks. Vertical handovers [6] and horizontal han-dovers, which can be qualified as hard or soft handovers, have been proposed. Theyrespectively allow the switching of the ongoing network connection from one wirelessinterface to another (e.g., handover from an 802.11b network into a GPRS network) andthe switching between two networks that use the same network technology and inter-face. With a hard handover mechanism, a mobile client can be connected with only oneaccess point at the same time, while a soft handover mechanism allows to keep two ormore connections with different access points. To the best of our knowledge, none ofthese handover solutions considers the issues inherent in the ICHNs.

3 Service provision in ICHNs

Three main issues must be overcome in order to efficiently provide nomadic peoplewith services in ICHNs, namely the discovery and the invocation of services usingopportunistic communications and the design of a handover mechanism in order tooffer a service access continuity to the mobile clients. In the first part of this section, wepresent different types of infrastructure of infostations, and we show how a handovermechanism should work in ICHNs. In the second part, we describe the service discoveryand invocation processes in such networks.

3.1 Handover overview and infostation infrastructures

(a) Architecture example 1. (b) Architecture example 2.

Fig. 2: Infrastuctures of infostations.

The infrastructure part of an ICHN can present various topologies (mesh, bus, etc.),and the service repositories can be organized in a centralized or distributed manner.For example, the infostations can provide the services themselves and can act as ser-vice repositories (see Figure 2a), or can simply act as gateways for other providers that


register their services within a centralized repository (see Figure 2b). Other kinds of ar-chitectures can obviously be considered. In the remainder of this paper we will assume,without loss of generality, that a service is provided directly by an infostation or viaanother one.

In order to provide mobile clients with an enhanced service access, the infosta-tions must estimate the ”quality” of the discontinuous/disconnected paths (DPs) be-tween themselves and the clients that require a service, must compare their estimationswith those computed by the other infostations, and, if necessary, must update their rout-ing table according to these new estimations.

In the handover solution we have devised, these estimations are obtained by theinfostations by processing the pieces of information stored in the service invocation re-quests they receive, such as the date of emission, the lifetime, the location of the client,etc. The computation algorithm and the properties we consider are detailed in Section 4.This handover solution works as follows: When the infostations receive an invocationrequest from a new client, or when they compute an estimation that is better than theprevious estimation they have in their routing table, they update their routing informa-tion and exchange summary vectors with the other infostations in order to allow them toupdate their own routing table in turn. The infostations are likely to not receive requestsfrom a given client during a long period, because this one has moved away, has becameisolated, or has been simply switched off. Thus, the information about this client mustno longer be stored in the routing table of the infostations. So as to cope with this issueand to maintain only the recent connections with mobile clients in the tables, we assigna date of computation and a lifetime to each entry. All the infostations thus share thesame perception of the infostation(s) that must forward the responses to a given client.In some situations, two (or more) infostations can approximately compute the same es-timations for a given client. These infostations are therefore considered as equivalent forthe service provision, and all of them should forward the service responses to the client,thus implementing a soft handover mechanism. In the remainder of this section, we de-scribe how this handover solution operates with the service discovery and invocationprocesses.

3.2 Service discovery

Service provision usually relies on three main operations: the discovery, the selectionand the invocation. In a wired network, the service discovery process is often based ona centralized approach: the providers register the services they offer within a registryand the clients can look up available services in this registry, and can obtain a refer-ence to the service they require. In an ICHN, the service discovery process cannot relyon a pure centralized approach since end-to-end routes between the mobile clients andthe fixed infostations do not exist permanently. Therefore, each client is responsible formaintaining its own perception of the services offered in the network, and for discover-ing these services either reactively by processing the unsolicited service advertisementsbroadcast periodically by service providers and/or proactively by broadcasting servicediscovery requests in the network and by processing the advertisements returned byproviders in response. This discovery process further helps mobile clients to select thepaths they must use to forward their requests toward the infostations. Indeed, in the


solution we propose the clients can process the pieces of information stored in the ser-vice advertisements they receive with an algorithm similar to that implemented in thehandover mechanism so as to evaluate the quality of the DPs and to select the best(s)DP(s), and thus to avoid a blindly forwarding process. Such a discovery can be achievedefficiently with OLFServ [8], which performs a geographically-constrained epidemicdissemination of both the service advertisements and service discovery requests.

3.3 Service invocation

A service invocation, during which a given client actually interacts with a provider, isusually performed using a unicast and destination-based communication model. Invok-ing a service in an ICHN basically consists in forwarding an invocation request toward agiven infostation, which in turn will process the request itself if it provides the requiredservice, or will forward the request to the infostation that provides this service. In thesolution we advocate, both service invocation messages and service response messagesare forwarded using source routing techniques in order to perform an efficient servicedelivery. Thus, while being forwarded, the messages are updated in order to include theIDs of the intermediate nodes, as well as the other properties that will allow to estimatethe quality of the discontinuous path. This list of IDs will then be used to compute thereverse route.

As mentioned previously, when a client requires a service for the first time, it es-timates the ”quality” of the DP between itself and an infostation based on the last ad-vertisements it receives. Then, it chooses the best reverse DP(s) that must be followedto forward an invocation. Sometimes several DPs can present approximately the samequality. When these DPs are considered as reliable enough, only one of these candidatepaths is selected (the best one). Otherwise, the messages will be forwarded followingeach distinct candidate path (i.e. following the paths that have no intersection betweentheir list of IDs of intermediate nodes). This DP (or these DPs) will be taken until thesource routing fails. When forwarding their responses toward the clients, the infosta-tions use the reverse route defined in the invocation message. When the source routingfails because an intermediate node is no longer reachable, the intermediate node thathas detected the failure will execute the same algorithm as the initial client, thus dy-namically updating the DP.

4 Handover mechanism for opportunistic computing

Handover decisions and route selections rely on the estimations of the ”quality” of theDPs. In the solution we propose, the DPs are characterized in terms of stability, ofdistance and of message propagation time. These metrics are defined below.

4.1 Message propagation time

The propagation time is an important metric in the service provision. It reflects thequality of service that is directly perceived by the end-users in terms of reactiveness. The


propagation time is computed either by the recipient or the destination of the message(i.e., by a mobile client or an infostation). The propagation time for a message m isgiven by pt(m, t) = t−m[de], where t is the date of reception of message m, m[de] isthe date of emission of message m.

4.2 Distance

We consider two different expressions of the notion of distance: a geographical dis-tance and an estimation of the physical distance based on the number of hops betweena source node and an infostation. The geographical distance between a client and aninfostation is given by:

d′(m) = R× arccos(sin(m[lat])× sin(latI)+ cos(m[lat])× cos(latI)× cos(m[lon]− lonI))

Where, R = 6378.137m, and the latitude and the longitude of the infostation and theclient are respectively defined in radians by (latI , lonI) and (m[lat],m[lon]).

For obvious reasons of energy consumption, nomadic people activate the GPS re-ceiver of their handheld devices only episodically. In order to cope with this issue, weuse another estimation of the distance based on the number of hops between a clientand an infostation. It must be noticed that, since the clients are mobiles and the linksare intermittent, a minimal number of hops between a client and an infostation doesnot guarantee a minimal geographical distance between these two entities. The estima-tion we propose therefore combines this number of hops with the message propagationtime in order to approximate the maximum distance between these two devices. Thisapproximation is defined as follows:

d”(m) = m[nh] ×CR+ s× (pt(m, t)−m[nh]×∆PT )

Where m[nh] is the number of hops for message m, ∆PT the delay of an immediateforwarding, CR the Wi-Fi communication range (typically 80 meters), and s the maxi-mum speed of movement of the node (typically 2 meters/seconds for a pedestrian).

The distance d(m) between a mobile client and an infostation is thus given byd(m) = d′(m) if the location properties are available, and is given by d(m) = d”(m)otherwise.

4.3 Path stability

The stability of a DP is another important metric because it reflects the ability to ef-ficiently forward a message to an infostation or to a mobile client using the sourcerouting technique, and the ability to recover an alternative path if the source routingfails. Consequently, we consider the number of neighbors of the intermediate nodesas an element of stability since it allows to take alternative paths if the source routingfails. Furthermore, this stability depends of several factors, such as the mobility of theintermediate nodes, their power budget, etc. Indeed, the devices are carried and used byhumans, and therefore can move freely or following social mobility patterns and canbe switched on/off for energy consumption purposes. In the current implementation of


our solution, we thus weight each estimation with the distance of a neighbor from theconsidered intermediate node if the locations, the speeds and the directions are known.Otherwise, we weight these estimations with the contact times that are simply definedby : ci = npi/np, where npi is the number of hello packets received from node i (i.e., thenumber of messages of presence sent by i), and where np is the number of hello packetsthe node i is expected to have sent since it has appeared in the vicinity of the currentnode. When the value of this property is equal (or close) to 1, node i is considered asa stable neighbor of the current node. At the opposite, a value close to 0 reflects thesporadic appearance of node i in the neighborhood of the current node. A lifetime is as-sociated with this value so as to consider only the last contacts between two nodes. Thepath stability estimation obtained locally (i.e., for a given intermediate node) is thus:

n

∑k=0

nsk, nsk =

{dk , i f location properties are availableck , otherwise

and dk =

{1 , i f distanceAt(locationk,sk,bkk,2×∆t)≤CR0 , otherwise

Where, locationk, sk and bk are respectively the current location and the speed ofmovement and the bearing of neighbor node k, ∆t the delay to forward a message toan infostation from the local node, CR the communication range of the local node, andck the contact times of node k. Function distanceAt() returns the distance between thelocal node and another node at a given time based on the location, the speed and thedirection of these two nodes. The path stability value is the minimum of the estimationsobtained along the path. It is thus defined as follows:

m[pathstatbility] = min(m[path stability],new estimation)

Where m[pathstability] is the stability of the path taken by message m. The functionthat returns the path stability is thus defined by s(m) = m[pathstability].

4.4 Handover algorithm

The handover algorithm aims at choosing the infostations that must forward the re-sponses to a given client based on the above presented metrics. Similarly, when theyhave the opportunity to forward their service invocation requests following differentDPs, the mobile clients apply a quite similar algorithm than that implemented in thehandover mechanism. In the remainder of this section, we focus only on the handoveralgorithm.

When an infostation receives an invocation request from a client, it estimates thequality of the path taken by the invocation request. Then it checks its routing table forthe previous estimations it has for this client. If it has no information about this client,it stores this estimation in its own routing table and sends to the other infostations ona multicast address a summary vector including the modifications it operates on itsrouting table so that they can propagate these modifications on their own routing tablein their turn (see Algorithm 1). If it finds some estimations for the considered client,


Algorithm 1 The section of the algorithm applied upon service invocation reception.Data:

R: the routing table m: the incoming invocation requestI: the current infostation D: the current dateF : the estimation function V : the summary vector

1: R ←R - {R{client=m[source] & infostation = I}}; T←R{client=m[source]}2: E ←F (m)3: if (T = Ø) then4: R ←R ∪ {m[source],I,D,E } ; V ← {add,{m[source],I,D,E }} ; send V5: else6: if (E ≥ max(T[estimation]) then7: if (E ≥ ΓE ) then8: R ←R ∪ {m[source],I,D,E } - T9: for all k ∈ T do

10: V ← V ∪ {remove,k}11: end for12: V ← V ∪ {add,{m[source],I,D,E }} ; send V13: else14: for all k ∈ T do15: if (k[estimation] + ∆E < E ) then16: R ←R - {k} ; V ← V ∪ {remove,k}17: end if18: end for19: V ← V ∪ {add,{m[source],I,D,E }} ; send V20: end if21: else22: if (E > max(T[estimation]) - ∆E ) then23: R ←R ∪ {m[source],I,D,E } ; V ← V ∪ {add,{m[source],I,D,E }} ; send V24: end if25: end if26: end if

the infostation checks if the new estimation is better than the previous ones. If so, itchecks again if this estimation is greater than ΓE . If so, it removes the older estimationsand only keeps the new one. ΓE is a parameter of the algorithm. When an estimation isgreater than ΓE , the path is considered as reliable and consequently it is not relevant toforward a message from two distinct infostations. If the new estimation is less than ΓE

and better than the previous ones, the infostation keeps only the better estimations thatare considered as equivalent (i.e. the estimations whose gap with the better estimationis less than ∆E ). A summary vector is sent to the other infostations in order to propagatethe modifications.

When they receive a summary vector, the infostation execute the simple algorithm 2,which consists in adding, removing or updating lines in the routing table.

F (m) = α× 1pt(m, t)

× s(m)× 1m[number o f hops]

× 1d(m)

The estimation of the ”quality” of the discontinuous paths is computed using thefunction defined above. This function aims at privileging the paths that offer a goodpropagation time and stability, as well as the infostations closer to the client. α is aparameter of the function that allows to obtain results greater than 1 (typically α can beequal to 1000).


Algorithm 2 The section of the algorithm applied upon summary vector reception.Data:

R: the routing table m: the incoming invocation requestI: the current infostation D: the current dateF : the estimation function V : the summary vector

1: for all k ∈ V do2: if k[action] = remove then3: R ←R - {k}4: end if5: if k[action] = add then6: R ←R ∪ {k}7: end if8: end for

5 Case study

In this section, we present the simulation results we have obtained for the handovermechanism described in previous sections, and we analyze the impact of this mecha-nism on the service delivery from the client point of view. The simulations have beenperformed on the OMNeT++ network simulator.

5.1 Environment

The environment we consider in these simulations is a square area of 1 km2 in whichwe have deployed 3 infostations. These infostations are connected together, and areseparated from each other of 400 m. Each of them provides a specific service. Theseservices are announced periodically (every 5 minutes) by all the infostations. They canbe discovered and invoked by pedestrians that move in this area using their handhelddevices. In these simulations, we consider two populations of pedestrians: the pedestri-ans that move following a random way point mobility model, and the pedestrians thatmove following predefined paths and that can exhibit their location. These pedestriansmove at a speed between 0.5 and 2 m/s. In our simulations, 30 % of the mobile devicesact as clients of the above-mentioned services, whereas the others only act as interme-diate nodes. After discovering the services they are looking for, the clients invoke theseservices every 3 minutes. They sent a maximum of 10 requests during the simulations.In our experiments, we have assigned to all the messages a lifetime of 10 minutes anda maximum number of hops of 10. The communication range of both mobile devicesand infostations varies from 60 to 80 m. In our simulations, we have considered succes-sively 50, 100, 200 and 300 pedestrians. All these parameters are defined so as to reflectas well as possible the behavior of humans that use their mobile phone when they arestrolling in a city.

5.2 Simulation results

The objective of these experiments is to measure the impact of our handover solutionon the service delivery in various configurations. For that, we focus especially on twovalues that reflect the quality of service that is perceived by the end-users (the ratio and


delay of service delivery), as well as on a value that shows the efficiency of the solution(the number of messages that are sent by all the nodes in the network throughout thewhole simulation period). The delivery ratio is the percentage of successful service in-vocation (i.e., the number of invocations for which a client node receives their responsefrom an infostation), while the service delivery delay is the time needed to forward aninvocation message toward the appropriate infostation, as well as to forward the re-sponse to the client. We compare the performance of our solution with the EpidemicRouting protocol [15]. In epidemic routing, messages are flooded in the network andstored by all available neighbor nodes as a result of summary vector exchanges, thusmaximizing the message delivery rate and minimizing message propagation latency.The first copy of a given service invocation request received by an infostation (or thefirst copy a given service response received by a client) has therefore followed the paththat offers the shortest delivery delay. Moreover, since the responses are disseminatedby all the nodes, including the infostations, no handover mechanism is required withthis protocol. In this context, the epidemic routing protocol appears as a good candi-date to evaluate the efficiency of our solution, even if no precautions are taken in thisprotocol to limit the number of messages that are disseminated.

Figures 3a, 3b and 3c present the simulation results we have obtained. One canobserve that our solution offers a better service delivery in terms of ratio and delaysthan the epidemic routing protocol, while reducing drastically the number of messagesthat are forwarded in the network, especially when the number of nodes increases. Thedelivery delays and service delivery ratios are often better with our proposal becausethe messages are forwarded using source routing techniques coupled with the handovermechanism resulting in the intervention of the infostation closest to the client, whilewith the epidemic routing protocol the messages are forwarded after the summary vec-tor exchanges. Due to this short additional latency in the message forwarding, somecommunication disruptions can occur in certain situations, thus reducing the opportu-nities to forward the messages. This difference is more observable when the number ofdevices is low because it is more difficult to find another intermediate node. Further-more, when the number of nodes increases in the network, the service delivery ratioincreases while the service delivery delay decreases. Indeed, as we notice, when havingfew nodes in the network, the satisfaction ratio of both protocols is almost the same.This observation is coherent with what is expected, because more good carriers can befound among a large set of neighbors, thus reducing the number of disruptions and thedisconnection times in the routes.

6 Conclusion

In this paper, we have presented a new soft handover solution suited for the serviceprovision in intermittently connected hybrid networks. This solution provides nomadicpeople with an enhanced service access by selecting the most appropriate discontin-uous path(s) between the clients and the infostations. The paths are characterized by


0

10

20

30

40

50

60

70

80

90

100

100 200 300

Perc

enta

ge

Number of nodes

Service provision with handoverEpidemic service provision

(a) Service delivery ratio.

0

1

2

3

4

5

6

7

8

100 200 300

tim

e (

min

tues)

Number of nodes


(b) Service delivery delay.

0

1000

2000

3000

4000

50 100 150 200 250 300

Nu

mb

er

of

me

ssa

ge

s

Number of nodes


(c) Network load.

Fig. 3: Simulation results.

three metrics, namely their stability, the propagation time they offer and their length.Furthermore, both the mobility and the number of neighbors of intermediate nodes aretaken into account in the stability estimation.

In the future, we plan to consider new kinds of properties such as the power budgetof the mobile devices. Finally, we wish to improve our handover solution by consid-ering the successive contacts of a mobile device with infostations, so as to predict itsdestination without location information and thus its next contact with an infostation.Consequently, allowing us to forward the responses in advance through this infostation.

References

1. Chiara Boldrini, Marco Conti, Iacopo Iacopini, and Andrea Passarella. Hibop: a historybased routing protocol for opportunistic networks. In Marco Conti, editor, Proc. IEEE In-


ternational Symposium on a World of Wireless, Mobile and Multimedia Networks WoWMoM2007, pages 1–12, 2007.

2. Marco Conti and Mohan Kumar. Opportunities in Opportunistic Computing. Computer,43:42–50, 2010.

3. Bo Han, Pan Hui, V.S. Anil Kumar, Madhav V. Marathe, Guanhong Pei, and Aravind Srini-vasan. Cellular traffic offloading through opportunistic communications: a case study. InProceedings of the 5th ACM workshop on Challenged networks, CHANTS ’10, pages 31–38,New York, NY, USA, 2010. ACM.

4. Pan Hui, Anders Lindgren, and Jon Crowcroft. Empirical evaluation of hybrid opportunisticnetworks. In Communication Systems and Networks and Workshops, 2009. COMSNETS2009. First International, pages 1–10, Bangalore,India, January 2009.

5. Stratis Ioannidis, Augustin Chaintreau, and Augustin Massoulie. Optimal and scalable dis-tribution of content updates over a mobile social network. In IEEE INFOCOM, Rio deJaneiro,Brazil, 2009.

6. Meriem Kassar, Brigitte Kervella, and Guy Pujolle. An overview of vertical han-dover decision strategies in heterogeneous wireless networks. Computer Communication,31(10):2607–2620, June 2008.

7. Nicolas Le Sommer, Salma Ben Sassi, Frédéric Guidec, and Yves Mahéo. A MiddlewareSupport for Location-Based Service Discovery and Invocation in Disconnected MANETs.Studia Informatica Universalis, 8(3):71–97, September 2010.

8. Nicolas Le Sommer and Yves Mahéo. OLFServ: an Opportunistic and Location-Aware For-warding Protocol for Service Delivery in Disconnected MANETs. In 5th International Con-ference on Mobile Ubiquitous Computing, Systems, Services and Technologies (Ubicomm2011), pages 115–122, Lisbon, Portugal, November 2011. Xpert Publishing Services.

9. Anders Lindgren, Avri Doria, and Olov Schelén. Probabilistic routing in intermittently con-nected networks. In Proceedings of The First International Workshop on Service Assurancewith Partial and Intermittent Resources (SAPIR 2004), August 2004.

10. Yves Mahéo and Romeo Said. Service Invocation over Content-Based Communication inDisconnected Mobile Ad Hoc Networks. In 24th International Conference on AdvancedInformation Networking and Applications (AINA’10), pages 503–510, Perth, Australia, April2010. IEEE CS.

11. Ali Makke, Nicolas Le Sommer, and Yves Mahéo. TAO: A Time-Aware Opportunistic Rout-ing Protocol for Service Invocation in Intermittently Connected Networks. In 8th Interna-tional Conference on Wireless and Mobile Communications (ICWMC 2012), pages 118–123,Venice, Italy, June 2012. Xpert Publishing Services.

12. Mirco Musolesi and Cecilia Mascolo. CAR: Context-Aware Adaptive Routing for DelayTolerant Mobile Networks. IEEE Transactions on Mobile Computing, 8(2):246–260, 2009.

13. Hoang Anh Nguyen, Silvia Giordano, and Alessandro Puiatti. Probabilistic routing protocolfor intermittently connected mobile ad hoc network (propicman). In IEEE InternationalSymposium on a World of Wireless, Mobile and Multimedia Networks (WOWMOM 2007),AOC Workshop, pages 1–6. IEEE Explore, jun 2007.

14. Luciana Pelusi, Andrea Passarella, and Marco Conti. Opportunistic Networking: Data For-warding in Disconnected Mobile Ad Hoc Networks. IEEE Communications Magazine,November 2006.

15. Amin Vahdat and David Becker. Epidemic Routing for Partially Connected Ad Hoc Net-works. Technical report, Duke University, April 2000.

16. John Whitbeck, Yoann Lopez, Jeremie Leguay, Vania Conan, and Marcelo Dias de Amorim.Relieving the Wireless Infrastructure: When Opportunistic Networks Meet Guaranteed De-lays. In 13th IEEE International Symposium on a World of Wireless Mobile and MultimediaNetworks (WoWMoM 2012), pages 1–10, San Francisco, California, USA, June 2011.

Documents

A Soft Handover for Service Delivery in Intermittently ... · A Soft Handover for Service Delivery in Intermittently Connected Hybrid Networks 3 according to the quality of the multi-hop